How should I ethically approach user password storage for later plaintext retrieval?

How about taking another approach or angle at this problem? Ask why the password is required to be in plaintext: if it's so that the user can retrieve the password, then strictly speaking you don't really need to retrieve the password they set (they don't remember what it is anyway), you need to be able to give them a password they can use.

Think about it: if the user needs to retrieve the password, it's because they've forgotten it. In which case a new password is just as good as the old one. But, one of the drawbacks of common password reset mechanisms used today is that the generated passwords produced in a reset operation are generally a bunch of random characters, so they're difficult for the user to simply type in correctly unless they copy-n-paste. That can be a problem for less savvy computer users.

One way around that problem is to provide auto-generated passwords that are more or less natural language text. While natural language strings might not have the entropy that a string of random characters of the same length has, there's nothing that says your auto-generated password needs to have only 8 (or 10 or 12) characters. Get a high-entropy auto-generated passphrase by stringing together several random words (leave a space between them, so they're still recognizable and typeable by anyone who can read). Six random words of varying length are probably easier to type correctly and with confidence than 10 random characters, and they can have a higher entropy as well. For example, the entropy of a 10 character password drawn randomly from uppercase, lowercase, digits and 10 punctuation symbols (for a total of 72 valid symbols) would have an entropy of 61.7 bits. Using a dictionary of 7776 words (as Diceware uses) which could be randomly selected for a six word passphrase, the passphrase would have an entropy of 77.4 bits. See the Diceware FAQ for more info.

• a passphrase with about 77 bits of entropy: "admit prose flare table acute flair"

• a password with about 74 bits of entropy: "K:&$R^tt~qkD"

I know I'd prefer typing the phrase, and with copy-n-paste, the phrase is no less easy to use that the password either, so no loss there. Of course if your website (or whatever the protected asset is) doesn't need 77 bits of entropy for an auto-generated passphrase, generate fewer words (which I'm sure your users would appreciate).

I understand the arguments that there are password protected assets that really don't have a high level of value, so the breach of a password might not be the end of the world. For example, I probably wouldn't care if 80% of the passwords I use on various websites was breached: all that could happen is a someone spamming or posting under my name for a while. That wouldn't be great, but it's not like they'd be breaking into my bank account. However, given the fact that many people use the same password for their web forum sites as they do for their bank accounts (and probably national security databases), I think it would be best to handle even those 'low-value' passwords as non-recoverable.

Imagine someone has commissioned a large building to be built - a bar, let's say - and the following conversation takes place:

Architect: For a building of this size and capacity, you will need fire exits here, here, and here.
Client: No, that's too complicated and expensive to maintain, I don't want any side doors or back doors.
Architect: Sir, fire exits are not optional, they are required as per the city's fire code.
Client: I'm not paying you to argue. Just do what I asked.

Does the architect then ask how to ethically build this building without fire exits?

In the building and engineering industry, the conversation is most likely to end like this:

Architect: This building cannot be built without fire exits. You can go to any other licensed professional and he will tell you the same thing. I'm leaving now; call me back when you are ready to cooperate.

Computer programming may not be a licensed profession, but people often seem to wonder why our profession doesn't get the same respect as a civil or mechanical engineer - well, look no further. Those professions, when handed garbage (or outright dangerous) requirements, will simply refuse. They know it is not an excuse to say, "well, I did my best, but he insisted, and I've gotta do what he says." They could lose their license for that excuse.

I don't know whether or not you or your clients are part of any publicly-traded company, but storing passwords in any recoverable form would cause you to to fail several different types of security audits. The issue is not how difficult it would be for some "hacker" who got access to your database to recover the passwords. The vast majority of security threats are internal. What you need to protect against is some disgruntled employee walking off with all the passwords and selling them to the highest bidder. Using asymmetrical encryption and storing the private key in a separate database does absolutely nothing to prevent this scenario; there's always going to be someone with access to the private database, and that's a serious security risk.

There is no ethical or responsible way to store passwords in a recoverable form. Period.

You could encrypt the password + a salt with a public key. For logins just check if the stored value equals the value calculated from the user input + salt. If there comes a time, when the password needs to be restored in plaintext, you can decrypt manually or semi-automatically with the private key. The private key may be stored elsewhere and may additionally be encrypted symmetrically (which will need a human interaction to decrypt the password then).

I think this is actually kind of similar to the way the Windows Recovery Agent works.

• Passwords are stored encrypted
• People can login without decrypting to plaintext
• Passwords can be recovered to plaintext, but only with a private key, that can be stored outside the system (in a bank safe, if you want to).